You jest, but that kind of thinking actually happened, once. Team B argued that the Soviet Union had developed a new submarine detection system that didn't depend on sound. When faced with the fact that nobody had found anything like it, they argued that this only proved the point: since the detection system didn't depend on sound, it could not itself be detected.
GPGPU and coding for 216-core systems. Engineering peer-to-peer networks that stay robust in the face of attackers tinkering with the protocol. Quantum computers if you focus on the science, huge datacenters if you focus on the engineering.
That was sort of what they tried in the US with Eldred v. Ashcroft, but it failed. The "nice" thing about doing copyright extensions with a finite creep each time is that it stays de jure finite while it is de facto infinite, so the people doing it never have to show their true intent as they stay within the letter of the law.
That's silly, but I guess that's why it's called politics. Different classes have different costs and requirements, and so differentiation already exists: by age, and in the case of acceleration, by ability as well.
Why does it have to be one size fits all? Some could go to these sorts of schools, some could go to those sorts of schools. You'd just have to find a way of telling the prospective autodidacts apart from those who would crash without school, and then apply the right tool for the job.
Broadcast page, as has been suggested elsewhere. The caller would send a nonce encrypted with the target number's public key, as well as a mixnet path (or path fragment, since it doesn't know where the other phone is), and all of that wrapped up in the required envelopes to get it to a random target tower. The network gets the message to that tower, and then the tower does a flood propagation system (or [insert advanced method here]) to broadcast the message. The responding phone decrypts the nonce and sends back a key encrypted by the caller's key, and a session can be initiated from there.
Actually making sure that users can't DoS the service this way is a bit harder, but should be possible with time servers. Say there's a time server, or network of servers, that provides a digitally signed timestamp. If a tower sees a timestamp from the caller, then later sees a timestamp within a certain delta (or no timestamp at all), it doesn't propagate the message.
It's also not like you can make the cell phone technology in any other way, location tracking will always be possible.
Sure you could. Use a mixmaster type network on top. The first tower will see some unspecified mobile phone doing something unknown (because the payload is inside the encrypted envelope), the second tower will see something either encrypted or decrypted coming from the first tower, and so on down. Communication in the other direction would be done by the initiator specifying which tower he's using, then the other person creates a mixnet route ending at that tower. The lag might suck, but it would be *possible*, and without traffic analysis, the first tower could only track unspecified mobile phones (no IDs or anything).
Then tell me why the Scandinavian countries (or Japan for that matter), with their strong state systems, rank better on the Transparency International corruption list than does the United States; or why Yeltsin-era Russia, with no holds barred capitalism after the destruction of the Soviet planning agencies, was so corrupt.
I think their (claimed) reasoning is that too many articles leaves too much exposed to vandals. If there are ten trillion articles, someone could slip into one of them and replace the text with OMG U SUXORZZZ!, and later, when that article becomes important, the error is exposed for everybody to see. It's worse with subtle edits: if someone alters Joe Random Local Politician's page to claim he had terrorist ties, that could lie dormant until local papers search Wikipedia when he's up for re-election.
Kinda tenuous, but I *think* that's their claimed reasoning.
The 3-SAT problem really is a problem saying "given this circuit of boolean gates, determine if there are any inputs you can set to make the final output gate signal TRUE". So consider you want to make a computer that implements the factor guessing game: you give it a composite number and a possible factor, and it shows a light if the factor you input is a factor of the composite number. It's possible to build that kind of computer because it's easy to check if the factor divides the composite: just divide and see if you get a remainder.
But now consider that the special purpose computer you built is really just a bunch of boolean gates. The program is fixed, so it can be modeled as a lot of gates, too. So you write down a description of this machine, where the input gates correspond to the composite and factor, and the final output gate is TRUE if the light would go on, otherwise FALSE. Then you fix the composite so they're no longer inputs - e.g. if you want the composite to be 3, you force the two first input gates high.
Now you have reduced factoring this number to finding input gate values so that the output gate returns TRUE... and that's what 3-SAT does. You dump the whole thing through the solver and it tells you which input gates you need to set; then you divide the composite by the number those gates represent, readjust the set of gates, and repeat.
If P == NP, any puzzle where you can ask if you've got the right answer and get a yes/no quickly can be solved quickly, too, and if there's a "best" solution, it can also be found quickly (through binary search).
Now imagine the puzzle is "find the most efficient antenna", or "find the most aerodynamic car frame that has enough space inside", or "find the shortest description of this stock market data", and you start to get at the power of the thing. Forget public key decryption -- a large swathe of engineering (and logistics) problems can be solved by just asking for a solution. Theorem proving can be automated, even.
(For pedantics: yes, I know "polytime" does not necessarily correspond to "quickly" -- but now try to find an n^1000 algorithm in the wild. Also, "shortest description of this stock market data" may be PSPACE since the decompression algorithm could run forever, but I assume you'd want to get the answer reasonably soon.)
Planets orbiting in regions which we identify as a star's "habitable" zone are potential locations to establish colonies.
I imagine it would be much easier to just build a Bernal sphere or O'Neill cylinder than to physically go to another world. For discovery purposes, nothing beats exoplanets, but for colonization, space stations are cheaper.
You have either missed or sidestepped my point. I do not dispute that non-transitivities exist; I discussed them myself. My point was that in the majority of cases, a non-transitive result is not "ambiguous" at all.
I thought you were disputing the non-transitivities themselves, saying they were simply a consequence of how Condorcet methods processed the data; but you were saying they exist but are not ambiguous. However, when faced with a non-transitivity, a method has to make a choice since its output is transitive - either a winner, or a ranking of the candidates. Thus it has to project the space that includes non-transitivity onto one that doesn't. My point was that IRV projects in one way and Condorcet methods in another. Since each single-method system (Condorcet or otherwise) acts according to a consistent logic that is defined for the whole space, there's no reason to say, out of hand, that Condorcet systems are questionable but IRV is not -- unless the question is whether Condorcet itself is desirable.
And in THAT context -- ambiguities caused by removal of candidates in the middle of the process -- it might be justified to call them "ambiguities". But that is only a small subset of the non-transitivities that can occur in voting systems.
The results do not involve elimination. What Tideman is saying is that if you run 1000 3-candidate elections, you'd see cycles in about 10 of them, and if you run 1000 15-candidate elections, you'd see cycles in about 91 of them.
I will concede that in certain ways Shulze beats out IRV. But I will also say that Shulze is likely to be a much harder sell to the voting public. It allows freer choice but many people may not understand the implications of their choices, or how the votes are actually tallied.
True, I grant that Schulze's complexity is a problem. It might be better to rely on its precedence (in that it has been used in many organizations without much trouble), or phrase it in terms of repeatedly finding the group of candidates that are not beaten by any outside the group, and eliminating the candidate that has the least victory margin. The latter approach was used in the planning stage of an attempt to introduce Schulze to Washington elections in 2006 (see the discussion group, although it's dead now). Unfortunately, the state representative didn't get re-elected and so he couldn't propose it.
One might also use Ranked Pairs, which consists of sorting the pairwise victories by strength and going down the list, adding one-on-one preferences except when they contradict earlier ones. Ranked pairs has simplicity, Schulze has precedence, which is more important, I don't know.
Even Approval would avoid the oddities of IRV. If you're going for incrementalism, it's probably the easiest change to make: just count overvotes.
Without going into a lot of detail, IRV minimizes the larger concerns regarding Arrow. Whether it does so in ways that are "better" than Shulze, I cannot say at this time. I have not seen an actual comparison of the two in that context.
There are two ways to do such a comparison in a neutral manner.
The first is by criteria. A method passes a criterion if it always elects consistent with that criterion. For instance, a method passes the Condorcet criterion if it always elects the Condorcet winner when he exists. The Wikipedia page says that IRV "eliminates vote splitting, reducing concerns about tactical voting and strategic nomination". The criterion that mirrors this is called clone independence. A method passes clone independence if making duplicate candidates (that voters rank next to each other, but not necessarily in the same order) doesn't alter the outcome. Methods that split votes wou
AFAIK, all systems that fully satisfy the Condorcet criteria fail to account for the "circular ambiguities", without altering the rules in ways that are questionable at best.
That's not really true. I think an analogy to the majority criterion (if someone has a majority of first preference votes, he should win), would clarify voting method criteria. It would be wrong to say that a method electing a majority-preferred candidate when he exists is "ambiguous" because the majority criterion doesn't tell you what the method will do if there is no majority favorite. Methods like Plurality, IRV, pretty much any you can name, satisfy the majority criterion, but how they act when there is no majority - which is most of the case when there are more than two candidates - differ.
So too is the case with Condorcet methods. Saying that a method passes the Condorcet criterion simply means that if there is a Condorcet winner - a candidate who beats every other one-on-one - the method will elect that candidate. What it does if that's not the case is not specified.
Why would "altering the rules in ways that are questionable at best" count against Condorcet but not against Majority? In both cases you have a method that follows its own logic, and the logic is designed so that it passes Condorcet (or Majority) when constrained by those criteria. IRV would be a one-method system with respect to majority favorite, and a contrived system like "Pick the majority favorite if there is one, otherwise pick the Range winner" would be a two-method system.
And I want to emphasize again that the "circular ambiguities" described are not actually "ambiguous" at all. That is an erroneous label. If the voting system in question is a valid reflection of voter preferences, then they represent real reflections of real voting preferences, and are therefore something that must be eliminated, not just an "ambiguity" that needs to be clarified.
Circular ambiguities is the very reason Arrow's theorem works. If there's a cycle and the method passes majority favorite, no matter who the method elects, you can remove one of the other candidates and have the method fail IIA. Circular ambiguities can also appear in yea-or-nay voting: proposal X is on the table, Y is proposed. A majority favors Y to X, so the motion to go to Y is passed. Then Z is proposed, a majority favors Z to Y, so the motion to go to Z is passed. Finally, X is proposed, a majority favors X to Z, so the motion to go to X is passed. Cycle. If the rules prohibit going back to X, then that means that the person who sets the agenda determines which proposal will win (which may be exploited through so-called "poison pill" legislation).
That suggests that the cycle is a true ambiguity among the voters. One majority is of one opinion, another is of another, a third is of a third. They overlap slightly, but the inconsistency is still there. Just like there may be cases when a majority doesn't prefer a single candidate, there may be cases when different majorities have different opinions. IRV makes a decision by using its logic, and Schulze makes a decision by using *its* logic. Neither IRV nor Schulze go "ooh, a cycle, what am I going to do?". The difference is that in one of these types of logic, the Condorcet winner will be elected when he exists.
And in fact such non-transitive outcomes are real events, reflecting real votes, that happen in the real world.
Only rarely. According to Nicolaus Tideman ("Collective Decisions and Voting: The Potential for Public Choice"), using a model fitted to the ranked elections he could find, the probability of a Condorcet winner is 99% when there are three candidates, 90% for 15 candidates, and even at a very impractical 30 candidates, still 78%.
In the two-method systems, you essentially abandon Condorcet in those situations and rely on something else. On the other hand, in those
Multi gig? Schulze, as well as most other Condorcet methods, are summable, which means you only need an amount of space polynomial with regards to the number of candidates to run the election. In Schulze's case, this is an N-by-N matrix that tells you how well a given candidate does against a given other candidate, kind of like round robin sports results. Assuming the election authority publishes the matrix, anybody can check it for himself if he so desires.
You might say that you have no proof that the matrix corresponds to the actual ballots; but you have no proof that the Plurality counts are accurate either, or the Approval counts for that matter. To have proof, you need transparency: each party checking that the count goes well, random recounts, that sort of thing; and you need that no matter if it's Condorcet, Plurality, or Approval.
Your are referring to only one form of runoff voting, where the person with the least votes for an office is eliminated in each round. There are other forms, not all of which eliminate in such a manner.
To my knowledge, the term "instant runoff voting" is only used for what's also called Hare's method or the Alternative Vote. Quoting Wikipedia: In the initial count, the first preference of each voter is counted and used to order the candidates. Each first preference counts as one vote for the appropriate candidate. Once all the first preferences are counted, if one candidate holds a majority, that candidate wins. Otherwise the candidate who holds the fewest first preferences is eliminated..
If you would eliminate the candidate that gets the most last-place votes, it would no longer be IRV, it would be Coomb's method. That method, too, exhibits the oddness that moving your candidate higher can make him lose; every method that eliminates one candidate at a time according to a weighted positional method (first place n points, second place k points..., last place p points) can do so.
As for "satisfy", I meant just that: studies of instant runoff versus other "simple" voting methods has shown that in practice, it results in choices that reflect the actual preferences of the most people.
Could you give me links? Without knowing what they mean by simple or by instant runoff, it's hard to say anything here. For instance, I would imagine Approval to satisfy more than IRV, or Minmax, where you pick the candidate whose worst one-on-one loss has the smallest margin of defeat, to do better as well, but perhaps the latter is no longer considered simple.
Systems that conform to the Condorcet criteria tend to have the problem that they do not account for non-transitivity of inequalities (which Wikipedia calls "circular ambiguities").
The Condorcet rule can be ambiguous, that much is correct. Yet all that means is that there's no "the" Condorcet method, you have to pick a method that conforms to the criterion and does whatever when there is no such winner. That can be as simple as adding a rule that you eliminate, among the bottom two in IRV, the one who loses one-on-one to the other; or it can be an entirely new rule, like Schulze or Ranked Pairs.
To argue against Condorcet because it's ambiguous would be like saying that any method that satisfies majority rule is suspect because majority rule doesn't tell you what to do if nobody got a majority.
Various forms of Instant Runoff minimize problems with non-transitivity and also Arrow's theorem, while retaining the strengths of systems like Condorcet. Therefore it is superior.
Instant Runoff Voting (at least the AV method, I don't know which others you mean) hides the non-transitivity. So do the Condorcet methods above - the methods don't act differently depending on whether or not there is ambiguity. Approval itself sidesteps the issue completely because it's not a ranked method.
Slashdot Mirror
You jest, but that kind of thinking actually happened, once. Team B argued that the Soviet Union had developed a new submarine detection system that didn't depend on sound. When faced with the fact that nobody had found anything like it, they argued that this only proved the point: since the detection system didn't depend on sound, it could not itself be detected.
On the other hand, Bernard Cohen considered plutonium less dangerous than caffeine, at least in terms of ingestion.
GPGPU and coding for 216-core systems. Engineering peer-to-peer networks that stay robust in the face of attackers tinkering with the protocol. Quantum computers if you focus on the science, huge datacenters if you focus on the engineering.
That was sort of what they tried in the US with Eldred v. Ashcroft, but it failed. The "nice" thing about doing copyright extensions with a finite creep each time is that it stays de jure finite while it is de facto infinite, so the people doing it never have to show their true intent as they stay within the letter of the law.
And like all proper internet addicts, they don't have care how they look either :p
They're going to drill a hole in the dark one's prison! This thing is the Bore! AIEEEE!
Or something.
Then why were no cobalt bombs made?
Didn't they have this sort of thing in Deus Ex?
And the wheel of reincarnation turns another step.
That's silly, but I guess that's why it's called politics. Different classes have different costs and requirements, and so differentiation already exists: by age, and in the case of acceleration, by ability as well.
Why does it have to be one size fits all? Some could go to these sorts of schools, some could go to those sorts of schools. You'd just have to find a way of telling the prospective autodidacts apart from those who would crash without school, and then apply the right tool for the job.
Broadcast page, as has been suggested elsewhere. The caller would send a nonce encrypted with the target number's public key, as well as a mixnet path (or path fragment, since it doesn't know where the other phone is), and all of that wrapped up in the required envelopes to get it to a random target tower. The network gets the message to that tower, and then the tower does a flood propagation system (or [insert advanced method here]) to broadcast the message. The responding phone decrypts the nonce and sends back a key encrypted by the caller's key, and a session can be initiated from there.
Actually making sure that users can't DoS the service this way is a bit harder, but should be possible with time servers. Say there's a time server, or network of servers, that provides a digitally signed timestamp. If a tower sees a timestamp from the caller, then later sees a timestamp within a certain delta (or no timestamp at all), it doesn't propagate the message.
It's also not like you can make the cell phone technology in any other way, location tracking will always be possible.
Sure you could. Use a mixmaster type network on top. The first tower will see some unspecified mobile phone doing something unknown (because the payload is inside the encrypted envelope), the second tower will see something either encrypted or decrypted coming from the first tower, and so on down. Communication in the other direction would be done by the initiator specifying which tower he's using, then the other person creates a mixnet route ending at that tower. The lag might suck, but it would be *possible*, and without traffic analysis, the first tower could only track unspecified mobile phones (no IDs or anything).
Then tell me why the Scandinavian countries (or Japan for that matter), with their strong state systems, rank better on the Transparency International corruption list than does the United States; or why Yeltsin-era Russia, with no holds barred capitalism after the destruction of the Soviet planning agencies, was so corrupt.
What you want is PACER, the only nuclear fusion plant whose construction is a "mere" engineering problem. Power by blowing up hydrogen bombs!
I think their (claimed) reasoning is that too many articles leaves too much exposed to vandals. If there are ten trillion articles, someone could slip into one of them and replace the text with OMG U SUXORZZZ!, and later, when that article becomes important, the error is exposed for everybody to see. It's worse with subtle edits: if someone alters Joe Random Local Politician's page to claim he had terrorist ties, that could lie dormant until local papers search Wikipedia when he's up for re-election.
Kinda tenuous, but I *think* that's their claimed reasoning.
The 3-SAT problem really is a problem saying "given this circuit of boolean gates, determine if there are any inputs you can set to make the final output gate signal TRUE". So consider you want to make a computer that implements the factor guessing game: you give it a composite number and a possible factor, and it shows a light if the factor you input is a factor of the composite number. It's possible to build that kind of computer because it's easy to check if the factor divides the composite: just divide and see if you get a remainder.
But now consider that the special purpose computer you built is really just a bunch of boolean gates. The program is fixed, so it can be modeled as a lot of gates, too. So you write down a description of this machine, where the input gates correspond to the composite and factor, and the final output gate is TRUE if the light would go on, otherwise FALSE. Then you fix the composite so they're no longer inputs - e.g. if you want the composite to be 3, you force the two first input gates high.
Now you have reduced factoring this number to finding input gate values so that the output gate returns TRUE... and that's what 3-SAT does. You dump the whole thing through the solver and it tells you which input gates you need to set; then you divide the composite by the number those gates represent, readjust the set of gates, and repeat.
If P == NP, any puzzle where you can ask if you've got the right answer and get a yes/no quickly can be solved quickly, too, and if there's a "best" solution, it can also be found quickly (through binary search).
Now imagine the puzzle is "find the most efficient antenna", or "find the most aerodynamic car frame that has enough space inside", or "find the shortest description of this stock market data", and you start to get at the power of the thing. Forget public key decryption -- a large swathe of engineering (and logistics) problems can be solved by just asking for a solution. Theorem proving can be automated, even.
(For pedantics: yes, I know "polytime" does not necessarily correspond to "quickly" -- but now try to find an n^1000 algorithm in the wild. Also, "shortest description of this stock market data" may be PSPACE since the decompression algorithm could run forever, but I assume you'd want to get the answer reasonably soon.)
It wouldn't have helped, as Voice of Cheney takes (took?) precedence.
I imagine it would be much easier to just build a Bernal sphere or O'Neill cylinder than to physically go to another world. For discovery purposes, nothing beats exoplanets, but for colonization, space stations are cheaper.
Get a connection from MyISP! /Up to/ 1 Gbps broadband!***
*** mean bandwidth 512/340 Kbps, 99th percentile 1Mbps/512Kbps.
I thought you were disputing the non-transitivities themselves, saying they were simply a consequence of how Condorcet methods processed the data; but you were saying they exist but are not ambiguous. However, when faced with a non-transitivity, a method has to make a choice since its output is transitive - either a winner, or a ranking of the candidates. Thus it has to project the space that includes non-transitivity onto one that doesn't. My point was that IRV projects in one way and Condorcet methods in another. Since each single-method system (Condorcet or otherwise) acts according to a consistent logic that is defined for the whole space, there's no reason to say, out of hand, that Condorcet systems are questionable but IRV is not -- unless the question is whether Condorcet itself is desirable.
The results do not involve elimination. What Tideman is saying is that if you run 1000 3-candidate elections, you'd see cycles in about 10 of them, and if you run 1000 15-candidate elections, you'd see cycles in about 91 of them.
True, I grant that Schulze's complexity is a problem. It might be better to rely on its precedence (in that it has been used in many organizations without much trouble), or phrase it in terms of repeatedly finding the group of candidates that are not beaten by any outside the group, and eliminating the candidate that has the least victory margin. The latter approach was used in the planning stage of an attempt to introduce Schulze to Washington elections in 2006 (see the discussion group, although it's dead now). Unfortunately, the state representative didn't get re-elected and so he couldn't propose it.
One might also use Ranked Pairs, which consists of sorting the pairwise victories by strength and going down the list, adding one-on-one preferences except when they contradict earlier ones. Ranked pairs has simplicity, Schulze has precedence, which is more important, I don't know.
Even Approval would avoid the oddities of IRV. If you're going for incrementalism, it's probably the easiest change to make: just count overvotes.
There are two ways to do such a comparison in a neutral manner.
The first is by criteria. A method passes a criterion if it always elects consistent with that criterion. For instance, a method passes the Condorcet criterion if it always elects the Condorcet winner when he exists. The Wikipedia page says that IRV "eliminates vote splitting, reducing concerns about tactical voting and strategic nomination". The criterion that mirrors this is called clone independence. A method passes clone independence if making duplicate candidates (that voters rank next to each other, but not necessarily in the same order) doesn't alter the outcome. Methods that split votes wou
That's not really true. I think an analogy to the majority criterion (if someone has a majority of first preference votes, he should win), would clarify voting method criteria. It would be wrong to say that a method electing a majority-preferred candidate when he exists is "ambiguous" because the majority criterion doesn't tell you what the method will do if there is no majority favorite. Methods like Plurality, IRV, pretty much any you can name, satisfy the majority criterion, but how they act when there is no majority - which is most of the case when there are more than two candidates - differ.
So too is the case with Condorcet methods. Saying that a method passes the Condorcet criterion simply means that if there is a Condorcet winner - a candidate who beats every other one-on-one - the method will elect that candidate. What it does if that's not the case is not specified.
Why would "altering the rules in ways that are questionable at best" count against Condorcet but not against Majority? In both cases you have a method that follows its own logic, and the logic is designed so that it passes Condorcet (or Majority) when constrained by those criteria. IRV would be a one-method system with respect to majority favorite, and a contrived system like "Pick the majority favorite if there is one, otherwise pick the Range winner" would be a two-method system.
Circular ambiguities is the very reason Arrow's theorem works. If there's a cycle and the method passes majority favorite, no matter who the method elects, you can remove one of the other candidates and have the method fail IIA. Circular ambiguities can also appear in yea-or-nay voting: proposal X is on the table, Y is proposed. A majority favors Y to X, so the motion to go to Y is passed. Then Z is proposed, a majority favors Z to Y, so the motion to go to Z is passed. Finally, X is proposed, a majority favors X to Z, so the motion to go to X is passed. Cycle. If the rules prohibit going back to X, then that means that the person who sets the agenda determines which proposal will win (which may be exploited through so-called "poison pill" legislation).
That suggests that the cycle is a true ambiguity among the voters. One majority is of one opinion, another is of another, a third is of a third. They overlap slightly, but the inconsistency is still there. Just like there may be cases when a majority doesn't prefer a single candidate, there may be cases when different majorities have different opinions. IRV makes a decision by using its logic, and Schulze makes a decision by using *its* logic. Neither IRV nor Schulze go "ooh, a cycle, what am I going to do?". The difference is that in one of these types of logic, the Condorcet winner will be elected when he exists.
Only rarely. According to Nicolaus Tideman ("Collective Decisions and Voting: The Potential for Public Choice"), using a model fitted to the ranked elections he could find, the probability of a Condorcet winner is 99% when there are three candidates, 90% for 15 candidates, and even at a very impractical 30 candidates, still 78%.
Multi gig? Schulze, as well as most other Condorcet methods, are summable, which means you only need an amount of space polynomial with regards to the number of candidates to run the election. In Schulze's case, this is an N-by-N matrix that tells you how well a given candidate does against a given other candidate, kind of like round robin sports results. Assuming the election authority publishes the matrix, anybody can check it for himself if he so desires.
You might say that you have no proof that the matrix corresponds to the actual ballots; but you have no proof that the Plurality counts are accurate either, or the Approval counts for that matter. To have proof, you need transparency: each party checking that the count goes well, random recounts, that sort of thing; and you need that no matter if it's Condorcet, Plurality, or Approval.
To my knowledge, the term "instant runoff voting" is only used for what's also called Hare's method or the Alternative Vote. Quoting Wikipedia: In the initial count, the first preference of each voter is counted and used to order the candidates. Each first preference counts as one vote for the appropriate candidate. Once all the first preferences are counted, if one candidate holds a majority, that candidate wins. Otherwise the candidate who holds the fewest first preferences is eliminated..
If you would eliminate the candidate that gets the most last-place votes, it would no longer be IRV, it would be Coomb's method. That method, too, exhibits the oddness that moving your candidate higher can make him lose; every method that eliminates one candidate at a time according to a weighted positional method (first place n points, second place k points..., last place p points) can do so.
Could you give me links? Without knowing what they mean by simple or by instant runoff, it's hard to say anything here. For instance, I would imagine Approval to satisfy more than IRV, or Minmax, where you pick the candidate whose worst one-on-one loss has the smallest margin of defeat, to do better as well, but perhaps the latter is no longer considered simple.
The Condorcet rule can be ambiguous, that much is correct. Yet all that means is that there's no "the" Condorcet method, you have to pick a method that conforms to the criterion and does whatever when there is no such winner. That can be as simple as adding a rule that you eliminate, among the bottom two in IRV, the one who loses one-on-one to the other; or it can be an entirely new rule, like Schulze or Ranked Pairs.
To argue against Condorcet because it's ambiguous would be like saying that any method that satisfies majority rule is suspect because majority rule doesn't tell you what to do if nobody got a majority.
Instant Runoff Voting (at least the AV method, I don't know which others you mean) hides the non-transitivity. So do the Condorcet methods above - the methods don't act differently depending on whether or not there is ambiguity. Approval itself sidesteps the issue completely because it's not a ranked method.